timberwolf 7.0 placement
DESCRIPTION
TimberWolf 7.0 Placement. Perform TimberWolf placement Based on the given standard cell placement Initial HPBB wirelength = 23. First Swap. Swap node b and e We shift node h : on the shorter side of the receiving row Node b included in nets { n 3 , n 9 }, and e in { n 1 , n 7 }. - PowerPoint PPT PresentationTRANSCRIPT
Practical Problems in VLSI Physical Design TimberWolf Placement (1/16)
TimberWolf 7.0 Placement Perform TimberWolf placement
Based on the given standard cell placement Initial HPBB wirelength = 23
Practical Problems in VLSI Physical Design TimberWolf Placement (2/16)
First Swap Swap node b and e
We shift node h: on the shorter side of the receiving row
Node b included in nets {n3, n9}, and e in {n1, n7}
Practical Problems in VLSI Physical Design TimberWolf Placement (3/16)
Computing ΔW ΔW = wirelength change from swap
Practical Problems in VLSI Physical Design TimberWolf Placement (4/16)
Estimating ΔWs ΔWs = wirelength change from shifting
h is shifted and included in n4 = {d,h,i} and n7 ={c,e,f,h,n}
h is on the right boundary of n4: gradient(h)++
h is not on any boundary of n7: no further change on gradient(h)
Practical Problems in VLSI Physical Design TimberWolf Placement (5/16)
Estimating ΔWs (cont)
Practical Problems in VLSI Physical Design TimberWolf Placement (6/16)
Accuracy of ΔWs Estimation How accurate is ΔWs estimation?
Node h is included in n4 = {d,h,i} and n7 ={c,e,f,h,n}
Real change is also 1: accurate estimation
Practical Problems in VLSI Physical Design TimberWolf Placement (7/16)
Estimation Model B Based on piecewise linear graph
Shifting h causes the wirelength of n4 to increase by 1 (19 to 20) and no change on n7 (stay at 28)
Practical Problems in VLSI Physical Design TimberWolf Placement (8/16)
Second Swap Swap node m and o
We shift node d and g: on the shorter side of the receiving row
Node m included in nets {n5, n9}, and o in {n2, n10}
Practical Problems in VLSI Physical Design TimberWolf Placement (9/16)
Computing ΔW ΔW = wirelength change from swap
Practical Problems in VLSI Physical Design TimberWolf Placement (10/16)
Estimating ΔWs Cell d and g are shifted
d is included in n4 = {d,h,i}, n6 ={d,k,j}, and n8 ={d,l}
d is on the right boundary of n6 and n8
So, gradient(d) = 2
Practical Problems in VLSI Physical Design TimberWolf Placement (11/16)
Estimating ΔWs (cont) Cell d and g are shifted
g is included in n1 = {a,e,g}, and n9 ={b,g,i,m}
g is on the right boundary of n1 and n9
So, gradient(g) = 2
Practical Problems in VLSI Physical Design TimberWolf Placement (12/16)
Estimating ΔWs (cont)
Practical Problems in VLSI Physical Design TimberWolf Placement (13/16)
Third Swap Swap node k and m
We shift node c: on the shorter side of the receiving row
Node k included in nets {n3, n6 , n10}, and m in {n5, n9}
Practical Problems in VLSI Physical Design TimberWolf Placement (14/16)
Computing ΔW ΔW = wirelength change from swap
Practical Problems in VLSI Physical Design TimberWolf Placement (15/16)
Estimating ΔWs Cell c is shifted
c is included in n3 = {b,c,k,n} and n7 ={c,e,f,h,n}
c is on the left boundary of n3
So, gradient(c) = −1
Practical Problems in VLSI Physical Design TimberWolf Placement (16/16)
Estimating ΔWs (cont)